Multi-element algorithm fusion-based anti-theft and anti-violation intelligent identification method
By using the method of multi-algorithm fusion, linear and nonlinear algorithms are used to generate feature space, and abnormal pattern recognition and latent feature correlation analysis are performed, which solves the problem of insufficient accuracy in the identification of electricity theft and illegal electricity use in existing technologies and improves the accuracy and adaptability of recognition.
Patent Information
- Application Number
- CN202510942459.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-09
- Publication Date
- 2025-10-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the existing anti-electricity theft and illegal electricity usage identification technologies, a single algorithm or a simple algorithm combination is difficult to accurately cover complex and highly disguised electricity theft or illegal behavior patterns. In particular, in the process of multi-algorithm fusion, the feature responses are inconsistent, forming implicit dependencies, which makes identification difficult.
A method based on multi-algorithm fusion is adopted to analyze the electricity consumption characteristic data through linear mapping algorithm and nonlinear sensitivity algorithm to generate linear and nonlinear feature spaces. Combined with abnormal pattern recognition and implicit feature correlation recognition, fusion correlation analysis is performed, and blind spot correction is used to improve recognition accuracy.
It improves the ability to identify illegal electricity use or electricity theft behaviors with strong disguise and weak feature expression, enhances the robustness and adaptability of recognition, and achieves comprehensive coverage of abnormal behaviors of multiple types and across feature dimensions.
Smart Images
Figure CN120804987A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of abnormal power consumption identification, and more particularly to an anti-spying and illegal power consumption identification method based on multi-algorithm fusion. BACKGROUND
[0002] In the existing anti-electricity stealing and illegal power consumption identification technology, a single algorithm or a simple algorithm combination is generally used to model and identify user power consumption behavior data.
[0003] With the increasing complexity of power consumption behavior and the intelligent disguise of electricity stealing means, the traditional method has obvious limitations. Especially in the multi-algorithm fusion process, different algorithms perform sensitivity analysis or nonlinear mapping on the same feature dimension, and due to inconsistent feature responses, potential implicit dependency relationships are formed, making it difficult to accurately cover some key electricity stealing or illegal behavior patterns. SUMMARY
[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an anti-spying and illegal power consumption identification method based on multi-algorithm fusion to solve the problems raised in the background art.
[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:
[0006] An anti-spying and illegal power consumption identification method based on multi-algorithm fusion, comprising the following steps:
[0007] S1: obtaining real-time monitoring data of user power consumption, and performing feature extraction on the real-time monitoring data of power consumption to obtain multi-dimensional power consumption feature data;
[0008] S2: performing linear mapping analysis and nonlinear sensitivity evaluation on the multi-dimensional power consumption feature data based on a linear mapping algorithm and a nonlinear sensitivity algorithm, respectively, to generate linear feature space and nonlinear feature space data;
[0009] S3: performing abnormal pattern identification processing on the linear feature space data to generate linear space abnormal features;
[0010] S4: performing implicit feature correlation identification processing on the nonlinear feature space data to generate nonlinear space coupling features;
[0011] S5: performing fusion correlation analysis on the linear space abnormal features and the nonlinear space coupling features to generate fusion feature blind spot area identification data;
[0012] S6: performing blind spot correction analysis based on the fusion feature blind spot area identification data to output illegal electricity stealing and illegal power consumption abnormal behavior identification results.
[0013] In a preferred embodiment, S1 comprises:
[0014] collecting power consumption load data and power consumption equipment running state data of a user in real time;
[0015] performing data preprocessing on the power consumption load data and the power consumption equipment running state data respectively, to remove abnormal data and interference data;
[0016] performing feature extraction processing on the power consumption load data and the power consumption equipment running state data after data preprocessing, to extract load fluctuation features, power consumption timing features and power consumption equipment running features, and to generate multi-dimensional power consumption feature data.
[0017] In a preferred embodiment, S2 comprises:
[0018] performing linear transformation processing on the multi-dimensional power consumption feature data based on a linear mapping algorithm, to generate linear feature space data reflecting differences in power consumption behavior features of the user;
[0019] performing sensitivity evaluation processing on the multi-dimensional power consumption feature data based on a nonlinear sensitivity algorithm, to determine the degree of nonlinear correlation between different power consumption features, and to generate nonlinear feature space data embodying implicit dependency relationships of power consumption features.
[0020] In a preferred embodiment, S3 comprises:
[0021] performing pattern extraction processing on the linear feature space data, to construct an initial abnormal behavior pattern set for discriminating power consumption behaviors of the user;
[0022] performing feature matching on the initial abnormal behavior pattern set, to identify a feature group deviating from normal behaviors by comparing each dimension feature in the linear feature space with predefined power consumption behavior standard parameters;
[0023] performing abnormal clustering based on the feature group deviating from normal behaviors, to distinguish different types of linear space abnormal behavior features, and to generate linear space abnormal feature data.
[0024] In a preferred embodiment, S4 comprises:
[0025] performing implicit feature correlation evaluation on the nonlinear feature space data, to calculate nonlinear correlation coefficients between dimension power consumption features, and to determine power consumption feature dimensions having implicit dependency relationships;
[0026] establishing a nonlinear feature dependency network model based on the power consumption feature dimensions having implicit dependency relationships, to evaluate the degree of implicit correlation between nodes in the nonlinear feature dependency network model;
[0027] based on the degree of implicit correlation between nodes, identifying a feature group exhibiting stable coupling relationships in the nonlinear feature space, and generating nonlinear space coupling feature data.
[0028] In a preferred embodiment, S5 comprises:
[0029] Fusion processing is performed on the linear space anomaly feature data and the nonlinear space coupling feature data to establish a fusion feature data mapping relationship.
[0030] Based on the fusion feature data mapping relationship, a feature correlation degree is calculated to determine the correlation strength between the linear space anomaly feature and the nonlinear space coupling feature.
[0031] According to the determined correlation strength, a fusion feature blind area is identified to determine the feature range that is not effectively covered in the fusion process, and fusion feature blind area identification data is generated.
[0032] In a preferred embodiment, S6 comprises:
[0033] Based on the fusion feature blind area identification data, blind spot feature reanalysis is performed to identify the anomaly feature type in the blind spot area.
[0034] According to the identified anomaly feature type in the blind spot area, weight adjustment is performed on the linear feature space data and the nonlinear feature space data, respectively, to generate blind spot corrected linear feature space data and nonlinear feature space data.
[0035] The blind spot corrected linear feature space data and the nonlinear feature space data are subjected to secondary fusion processing to output the illegal electricity stealing anomaly behavior identification result.
[0036] The technical effects and advantages of the present application, a multi-element algorithm fusion based anti-spying illegal electricity use intelligent identification method, are as follows:
[0037] By constructing a combination mechanism of linear mapping algorithm and nonlinear sensitivity algorithm, the modeling depth and feature expression ability of multi-dimensional electricity use feature data are effectively improved. Through anomaly pattern recognition in the linear feature space and implicit feature correlation recognition in the nonlinear feature space, multi-perspective capture of electricity use behavior anomalies is realized, and the coverage ability of multi-type, cross-feature dimension anomaly behavior is enhanced. Through fusion analysis and identification algorithm decision blind area, and combined with the blind spot correction mechanism, the feature space is dynamically optimized, the precision identification ability of disguised electricity use or electricity stealing behavior with weak feature expression is improved, and the identification robustness and adaptability are higher. BRIEF DESCRIPTION OF DRAWINGS
[0038] Figure 1 A schematic diagram of the present application, a multi-element algorithm fusion based anti-spying illegal electricity use intelligent identification method, is shown. DETAILED DESCRIPTION
[0039] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those ordinarily skilled in the art without creative effort belong to the scope of the present application.
[0040] Embodiments
[0041] Figure 1 A kind of anti-spying violation intelligent identification method based on multi-element algorithm fusion of the present application is given, which includes the following steps:
[0042] S1: obtaining the real-time monitoring data of user's electricity, and extracting features from the real-time monitoring data of electricity, to obtain multi-dimensional electricity feature data;
[0043] S2: based on linear mapping algorithm and nonlinear sensitivity algorithm, linear mapping analysis and nonlinear sensitivity evaluation are carried out on the multi-dimensional electricity feature data respectively, to generate linear feature space and nonlinear feature space data;
[0044] S3: linear feature space data is processed for abnormal pattern recognition, to generate linear space abnormal features;
[0045] S4: the nonlinear feature space data is processed for implicit feature correlation identification, to generate nonlinear space coupling features;
[0046] S5: linear space abnormal features and nonlinear space coupling features are fused and associated, to generate fusion feature blind spot area identification data;
[0047] S6: based on fusion feature blind spot area identification data, blind spot correction analysis is carried out, to output violation abnormal behavior identification result.
[0048] S1: obtaining the real-time monitoring data of user's electricity, and extracting features from the real-time monitoring data of electricity, to obtain multi-dimensional electricity feature data, including:
[0049] Real-time acquisition of user's electricity load data and electricity equipment operating state data;
[0050] Real-time collection of user power consumption load data refers to real-time collection of current, voltage, active power, reactive power, power factor, and power consumption of users in different time periods by using smart meters, load collection terminal devices, and other devices installed on the user side to obtain load information reflecting the actual power consumption characteristics of users. For example, a smart meter automatically collects and uploads user power consumption and power data to a data collection center at a fixed time interval, such as every 15 minutes, to achieve real-time continuous data acquisition, which can reflect the actual power consumption of users at different times and avoid misjudgment of user power consumption behavior due to data loss.
[0051] Real-time collection of user power consumption equipment operating state data refers to continuous monitoring and collection of operating parameters of main power consumption equipment at the user end. For example, real-time monitoring of operating state parameters of electrical equipment in user households or industrial scenarios, including but not limited to electric motors, air conditioners, refrigerators, heaters, and other equipment, such as device opening and closing states, current fluctuation during device operation, operating voltage changes, transient current characteristics during startup and shutdown, and other parameters. For example, when a user uses a household air conditioner, the startup, shutdown, and operation of the device will produce specific current and power variation characteristics. Real-time collection of device operating state data can provide detailed power consumption equipment operating information and enhance the accuracy and relevance of data collection.
[0052] Data preprocessing is performed on the power consumption load data and the power consumption equipment operating state data respectively to remove abnormal data and interference data.
[0053] Data preprocessing of power consumption load data refers to normalization, abnormal data filtering, and data smoothing processing of user load data to remove abnormal data and interference data caused by device failure, communication error, electromagnetic interference, or sudden abnormal behavior during data collection. For example, if the power data collected by a user at a certain moment suddenly appears an abnormal value far exceeding the user's rated capacity, the abnormal data needs to be identified and removed by setting a specific threshold; when the user's power consumption data has occasional jumps or burr phenomena, median filtering or moving average filtering can be used for data smoothing to ensure the continuity and accuracy of the obtained power consumption load data.
[0054] Data preprocessing of power consumption equipment operating state data refers to normalization, noise filtering, and data verification processing of power consumption equipment operating state data to remove interference data caused by device sensor jitter, environmental interference, or unstable signal transmission. For example, if the power consumption equipment state sensor has a temporary fault that causes abnormal fluctuations or distortion in the current monitored at a certain moment, threshold judgment or waveform analysis methods can be used to exclude abnormal data to ensure the authenticity and availability of the device operating state data.
[0055] The power consumption load data and the power consumption equipment operation state data after data preprocessing are subjected to feature extraction processing, and load fluctuation features, power consumption timing features and power consumption equipment operation features are extracted to generate multi-dimensional power consumption feature data.
[0056] The feature extraction processing on the preprocessed power consumption load data refers to extracting key features reflecting the differences in user power consumption behaviors according to the characteristic rules of load data changes. The load fluctuation features include, for example, the difference between the load peak value and the load valley value, the load intra-day fluctuation amplitude, the load change rate, the load stability index, etc., to represent the user load fluctuation features; the power consumption timing features include, for example, the user daily load curve features, the load change trend features, the power consumption periodicity features, the load timing mode features in different seasons or different power consumption modes, to reflect the timing regularity of user power consumption behaviors. The above features can reflect the mode differences of user power consumption.
[0057] The feature extraction processing on the power consumption equipment operation state data refers to extracting feature parameters reflecting the operation state of different types of equipment. For example, the duration feature of power consumption equipment operation, the transient current feature of equipment start and stop, the equipment operation stability feature, the current fluctuation frequency spectrum feature during operation, etc. The equipment operation features can be combined with the user load features to form a feature description of the real power consumption of the user.
[0058] The extracted load fluctuation features, power consumption timing features and power consumption equipment operation features are unified and fused to form multi-dimensional power consumption feature data, i.e., a multi-dimensional information set covering power consumption load behavior rules, power consumption equipment state rules, etc.
[0059] S2: linear mapping analysis and nonlinear sensitivity evaluation are performed on the multi-dimensional power consumption feature data based on linear mapping algorithm and nonlinear sensitivity algorithm to generate linear feature space and nonlinear feature space data, including:
[0060] Based on the linear mapping algorithm, linear transformation processing is performed on the multi-dimensional power consumption feature data to generate linear feature space data reflecting the differences in user power consumption behavior features;
[0061] The linear mapping algorithm is used to perform linear transformation processing on the multi-dimensional electricity consumption characteristic data, that is, linear operation is performed on the multi-dimensional electricity consumption characteristic data to obtain a characteristic space that can reflect the differences in electricity consumption behavior between users. The multi-dimensional electricity consumption characteristic data includes load fluctuation characteristics, electricity consumption timing characteristics, and electricity consumption equipment operation characteristics. The characteristic data is usually in the form of a multi-dimensional numerical matrix or high-dimensional data. The linear mapping algorithm, such as the principal component analysis method or the linear discriminant analysis method, is used to map the high-dimensional characteristic space data to a lower-dimensional space through a linear function, ensuring that the mapped characteristic space data retains the core difference characteristics of the original characteristic data and reduces the redundancy of the characteristic data.
[0062] For example, the principal component analysis can be used to perform linear mapping processing on the user electricity consumption load data. Covariance matrix calculation is performed on the collected multi-dimensional load fluctuation characteristics and electricity consumption timing characteristics to determine the correlation and variance contribution rate of different dimensional load characteristics. The principal component feature with the largest contribution rate is extracted to reduce the original multi-dimensional characteristic space to a low-dimensional linear space, significantly improving the data analysis efficiency. At the same time, the linearly transformed linear characteristic space data can effectively distinguish the differences in electricity consumption behavior between different users.
[0063] For example, linear mapping analysis is performed on the electricity consumption equipment operation characteristics of the user. Specifically, the covariance or correlation matrix between the equipment operation characteristic parameters is calculated to determine the importance of different equipment operation characteristics. Linear combination is used to compress multiple equipment characteristic dimensions to a few representative characteristic dimensions, so that the linearly mapped characteristic space can accurately reflect the difference characteristics of the operation state of different user equipment, thereby realizing the description of the user equipment operation behavior in a low-dimensional space.
[0064] The linear characteristic space data generated by the linear mapping algorithm can effectively reduce the complexity of the original multi-dimensional characteristic space data and reflect the main difference characteristics of the user electricity consumption behavior.
[0065] The sensitivity evaluation processing of the multi-dimensional electricity consumption characteristic data using the nonlinear sensitivity algorithm refers to using a sensitivity analysis method that can capture and analyze complex nonlinear relationships to evaluate the degree of nonlinearity and interaction between each characteristic dimension. The multi-dimensional electricity consumption characteristic data includes load fluctuation characteristics, electricity consumption timing characteristics, and electricity consumption equipment operation characteristics.
[0066] Specifically, the nonlinear sensitivity algorithm can use kernel function method, mutual information analysis, nonlinear correlation coefficient calculation and other nonlinear evaluation methods to calculate and evaluate the correlation degree between multi-dimensional power consumption feature data. For example, through kernel function sensitivity analysis, the feature data is mapped to a high-dimensional nonlinear feature space for calculation, and the nonlinear dependence relationship between the load fluctuation feature and the equipment operation feature is accurately evaluated.
[0067] The nonlinear feature space data obtained based on the nonlinear sensitivity algorithm can accurately identify the complex nonlinear relationship hidden between different features, and can also intuitively reflect the implicit dependence relationship between power consumption features.
[0068] Based on the nonlinear sensitivity algorithm, the sensitivity evaluation process is performed on multi-dimensional power consumption feature data to determine the nonlinear correlation degree between different power consumption features, and nonlinear feature space data reflecting the implicit dependence relationship between power consumption features is generated.
[0069] S3: performing abnormal pattern recognition processing on the linear feature space data to generate linear space abnormal features, including:
[0070] Performing pattern extraction processing on the linear feature space data to construct an initial abnormal behavior pattern set for discriminating user power consumption behavior;
[0071] The pattern extraction processing on the linear feature space data refers to analyzing the structure and regularity inside the data based on the linear feature space data using pattern extraction methods to extract initial feature combinations that can reflect user abnormal power consumption behavior. Statistical pattern recognition methods or feature threshold analysis methods are used. For example, set the normal range standards of different dimensional features in the linear feature space data, such as defining the load fluctuation amplitude feature and the power consumption daily load time sequence feature as the threshold range of normal power consumption behavior, respectively, compare and detect the user's linear feature space data one by one, if the load fluctuation feature, power consumption time sequence feature or equipment operation feature of the user exceeds the normal threshold, mark the corresponding feature combination as an initial abnormal behavior pattern, thereby constructing an initial abnormal behavior pattern set.
[0072] Performing feature matching on the initial abnormal behavior pattern set to identify feature groups that deviate from normal behavior by comparing the dimensional features in the linear feature space with the predefined power consumption behavior standard parameters;
[0073] Feature matching on the initial set of abnormal behavior patterns refers to one-by-one comparison of each abnormal behavior pattern in the initial set of abnormal behavior patterns with pre-defined standard power consumption behavior parameters to determine data in the user's linear feature space data that deviates from the standard parameters, thereby explicitly showing the feature difference of the initial abnormal behavior pattern. The pre-defined power consumption behavior standard parameters are normal power consumption feature standards obtained from a large amount of user historical data or experience statistics, such as typical load fluctuation characteristic values, normal power consumption timing curve characteristics, normal state characteristics of power consumption equipment, etc. of a certain industry or residential user group. When the features in the user's linear feature space data deviate from the power consumption behavior standard parameters, it can be judged that the user has abnormal behavior features.
[0074] For example, the standard feature range of the industrial user's power consumption load within a day is that the difference between the maximum and minimum load values is less than 20%, but the user's data shows that the load within a day fluctuates continuously for more than 50% after pattern extraction, which significantly deviates from the standard parameters, and the user is identified as having abnormal load fluctuation characteristics through feature matching analysis.
[0075] Abnormal clustering based on the features deviating from normal behavior, distinguishing different types of linear space abnormal behavior features, generating linear space abnormal feature data;
[0076] Abnormal clustering analysis based on the identified features deviating from normal behavior refers to classifying the identified abnormal features, and grouping features with similar or common abnormal feature performances into the same abnormal behavior category. The abnormal clustering analysis method can use methods such as hierarchical clustering, K-means clustering, density clustering, etc. to group and distinguish the identified features. For example, when the abnormal load fluctuation features of multiple users all show continuous abnormal increase at night, they can be clustered into one category; while another group of users have continuous abnormal load increase during the day, they can be classified into another category of abnormal features. Through the above methods, different types of abnormal power consumption behavior features can be defined and distinguished.
[0077] Specifically, a group of users show similar abnormal transient current features of power consumption equipment starting and stopping, and all deviate from normal equipment operation parameter features, which can be clustered into one category of equipment operation abnormal features; while another group of users show high load fluctuation features and have common timing patterns, such as abnormal load increase in abnormal time periods, which can be clustered into one category of load abnormal features. Through abnormal clustering, the abnormal behavior features in the linear space are classified and distinguished, which helps to define the feature range and boundary of each type of abnormal behavior.
[0078] After the clustering analysis is completed, linear space abnormal feature data is generated based on the abnormal feature categories obtained by clustering. The linear space abnormal feature data contains explicit and different types of abnormal power consumption behavior features, which facilitates fusion and correlation analysis with nonlinear space data, and more accurately identifies the types of abnormal power consumption behavior of users.
[0079] S4: Perform implicit feature correlation identification processing on the nonlinear feature space data to generate nonlinear space coupling features, including:
[0080] The implicit feature correlation of the nonlinear feature space data is evaluated, and the nonlinear correlation coefficients between the dimensions of the power consumption features are calculated to determine the power consumption feature dimensions that have implicit dependency relationships;
[0081] The implicit feature correlation of the nonlinear feature space data refers to analyzing the implicit correlation relationships between different dimensions of power consumption features based on nonlinear feature space data. The nonlinear feature space data contains nonlinear interaction relationships between load fluctuation features, power consumption timing features, and device operation features, and there are complex nonlinear dependency relationships between the features. Implicit feature correlation evaluation uses nonlinear statistical analysis techniques such as mutual information calculation, nonlinear correlation coefficient method, kernel density estimation method, etc. to analyze the degree of nonlinear correlation between feature dimensions and reveal the dependency relationships between features.
[0082] For example, the nonlinear correlation coefficient method is used to calculate the nonlinear correlation coefficients between the dimensions of power consumption features. Using methods such as Spearman rank correlation coefficient or distance correlation coefficient, the nonlinear correlation coefficients between load fluctuation features and power consumption device operation features, and between load timing features and device operation state features are calculated. If the correlation coefficient is higher than the set threshold, it indicates that there is a nonlinear implicit dependency relationship between the feature dimensions. For example, there is a nonlinear correlation phenomenon between the user's air conditioning device operation feature and the load fluctuation feature, i.e. the change in the device operation state does not simply cause a linear change in the load, but presents a complex nonlinear relationship. Nonlinear correlation coefficient calculation can effectively capture the nonlinear relationship and determine the feature dimensions that have nonlinear implicit dependency relationships.
[0083] Based on the power consumption feature dimensions that have implicit dependency relationships, a nonlinear feature dependency network model is established to evaluate the implicit correlation degree between nodes in the nonlinear feature dependency network model.
[0084] Establishing a nonlinear feature-dependency network model based on electricity consumption feature dimensions with implicit dependencies means using the identified electricity consumption feature dimensions with nonlinear implicit dependencies as nodes in the nonlinear feature-dependency network model, and constructing a network structure by establishing nonlinear association boundaries between nodes to reflect the nonlinear interdependencies between feature dimensions. The nonlinear feature-dependency network model can adopt network analysis methods, such as complex network model construction methods, to treat each identified feature dimension as a network node, and the degree of association between nodes is quantified using nonlinear correlation coefficient values or mutual information values. The nonlinear feature-dependency network model can not only intuitively reflect the degree of nonlinear association between feature dimensions, but also describe the complex dependency network structure between nodes.
[0085] Evaluating the degree of implicit connections between nodes in a nonlinear feature-dependent network model involves quantitatively analyzing the strength of node connections within the network model. By analyzing the weights of the associated edges between nodes in the network, we can determine the degree of connection between each feature node and other nodes. Ultimately, we generate quantitative results of node connection strength, which serve as the basis for identifying feature groups with stable coupling relationships.
[0086] Based on the implicit correlation degree between nodes, the feature groups showing stable coupling relationships in the nonlinear feature space are identified to generate nonlinear spatial coupling feature data;
[0087] Based on the quantification of the implicit correlation between nodes, identifying feature groups that exhibit stable coupling relationships in nonlinear feature spaces involves analyzing feature nodes with strong and long-term inter-node correlations to determine the characteristic dimensions that collectively constitute specific stable coupling relationships. Stable coupling relationship feature groups indicate that the nonlinear relationships between features are highly repeatable and stable, making them important signatures for identifying abnormal behavior in nonlinear spaces.
[0088] The above process ultimately generates nonlinear spatial coupling feature data. This data lists the stable feature coupling groups identified during the nonlinear feature space data analysis, recording and defining the implicit but stable dependencies between features. Nonlinear spatial coupling feature data can describe stable nonlinear association patterns between feature dimensions and provide a data foundation for fusion analysis of linear and nonlinear spatial features.
[0089] S5: Perform fusion correlation analysis on linear spatial anomaly features and nonlinear spatial coupling features to generate fusion feature blind spot area identification data, including:
[0090] Fusing linear space anomaly feature data with nonlinear space coupling feature data to establish a fusion feature data mapping relationship;
[0091] The fusion processing of linear space abnormal feature data and nonlinear space coupling feature data refers to comprehensively considering the linear space abnormal feature data and the nonlinear space coupling feature data, and realizing the effective fusion of the two kinds of feature data. The linear space abnormal feature data includes abnormal behavior feature categories extracted and clustered through linear mapping analysis, such as user load intra-day abnormal fluctuation amplitude feature, power consumption timing mode abnormal feature, equipment operation abnormal feature, etc. The abnormal features show the pattern of deviating from normal power consumption behavior in their respective linear dimensions. The nonlinear space coupling feature data includes stable feature coupling relationship features identified by nonlinear sensitivity analysis and feature dependent network model, such as the nonlinear correlation relationship between specific equipment operation state features and load fluctuation features. Through fusion processing, that is, through data level combination analysis and cross comparison, the correlation and difference between linear space abnormal features and nonlinear space coupling features are considered, and a fusion feature data mapping relationship is constructed. The establishment of the fusion feature data mapping relationship can adopt data fusion methods, such as feature level fusion, decision level fusion or association mapping fusion technology. For example, through feature level fusion, a unified data feature space can be constructed to unify the dimensions of linear abnormal feature data and nonlinear coupling feature data for coordinate mapping. Specifically, a mapping function is defined, such as a weighted feature mapping or a nonlinear transformation mapping method, to uniformly represent the linear abnormal feature data and the nonlinear coupling feature data in a unified space, so as to form a mapping relationship between the linear space abnormal feature data and the nonlinear space coupling feature data.
[0092] For example, when performing feature level fusion processing on the abnormal load fluctuation feature data of an industrial user, the load abnormal features in the linear space, such as the frequency and amplitude features of load fluctuation abnormality, and the load fluctuation and equipment operation stable coupling relationship features in the nonlinear space can be combined to construct a feature mapping relationship, so that the features of the two spaces are uniformly mapped into the same data matrix or feature space, so that the relationship structure between the linear space and the nonlinear space data can be observed, and a fusion feature data mapping relationship is formed.
[0093] Based on the fusion feature data mapping relationship, the feature correlation degree is calculated to determine the correlation strength between the linear space abnormal feature and the nonlinear space coupling feature;
[0094] Based on the fusion feature data mapping relationship, the feature correlation degree is calculated to determine the correlation strength between the linear space abnormal feature and the nonlinear space coupling feature;
[0095] Specifically, the mutual information value of each dimension feature in the fusion feature data space is calculated by joint mutual information analysis method, and the correlation strength between the abnormal load fluctuation features of the linear space and the coupling groups of the load fluctuation and equipment operation features of the nonlinear space is determined; the result of correlation strength calculation can be represented by a numerical value, such as the value of correlation coefficient, which directly reflects the correlation between the features and reveals the strength of the potential coupling relationship between the features.
[0096] For example, it is found by calculation that the correlation strength between the abnormal amplitude feature of load intra-day fluctuation in the linear feature space and the abnormal coupling feature of air conditioner operation duration in the nonlinear feature space is particularly high, which means that there is a very stable coupling relationship between the two groups of features.
[0097] According to the determined correlation strength, the fusion feature blind spot area is identified, the feature range not effectively covered in the fusion process is determined, and the fusion feature blind spot area identification data is generated;
[0098] Based on the feature correlation strength, the process of identifying the fusion feature blind spot area is to identify the feature data area that is not effectively covered in the fusion analysis process through the analysis result of the correlation degree of the features, that is, the fusion feature blind spot area. The fusion feature blind spot area may exist without explicit abnormal behavior features, which may cause possible identification omission.
[0099] Specifically, the correlation strength threshold can be set as a standard, and the actual value of the feature correlation strength is compared with the set correlation strength threshold to determine the blind spot area in the feature space that does not reach the effective coverage standard. For example, by setting the correlation strength threshold to 0.8, if the correlation strength calculation result of a feature group is less than 0.8, it means that the area is not effectively identified in the fusion process and should be included in the fusion feature blind spot area. Through the above method, the blind spot feature area existing in the fusion feature data can be effectively located.
[0100] The result of identifying the fusion feature blind spot area is presented in the form of fusion feature blind spot area identification data, which records and labels the identification coverage of each feature dimension or feature group in the feature fusion process, and is embodied as a data table or a marked matrix, which indicates the correlation strength value of each feature dimension in the fusion process and the position and range of the blind spot area that cannot be effectively covered.
[0101] S6: Based on the fusion feature blind spot area identification data, blind spot correction analysis is performed to output the illegal electricity stealing abnormal behavior identification result, including:
[0102] Based on the fusion feature blind spot area identification data, the blind spot feature is reanalyzed to identify the abnormal feature type in the blind spot area.
[0103] The process of blind spot feature reanalysis based on the fusion feature blind spot area identification data refers to reanalyzing the power consumption features in the blind spot area that is not effectively covered by using the fusion feature blind spot area identification data. The fusion feature blind spot area identification data embodies the information of the feature dimensions or data areas that are identified as blind spot areas in the feature fusion process.
[0104] The feature reanalysis of the blind spot area includes re-executing a higher-precision analysis method on the original linear abnormal feature data and nonlinear coupling feature data in the blind spot area, such as fine clustering analysis of local features, local feature sensitivity evaluation, and local feature threshold analysis. For example, more sensitive abnormal detection techniques, such as local anomaly factor analysis or isolation forest method, can be used to re-calculate and analyze the load abnormal fluctuation amplitude features or equipment operation state features in the blind spot area to determine the abnormal degree and abnormal behavior type of the features in the blind spot area.
[0105] For example, the correlation strength between the load fluctuation and the equipment operation feature of a certain industrial user in the fusion feature blind spot area identification data does not reach the preset standard, and there may be potential abnormal behavior that is not effectively covered by the initial identification. In the blind spot area, the user's load time series feature data and equipment operation transient current data are analyzed again, for example, the local anomaly factor method is used to analyze that there are high-frequency event features of short-time abnormal fluctuations in the area, and it is determined that the potential abnormal feature type in the blind spot area is the equipment frequent start abnormal behavior mode.
[0106] Through the blind spot feature reanalysis process, the abnormal behavior types missed in the fusion analysis process can be more accurately identified, thereby improving the completeness and accuracy of the abnormal behavior identification.
[0107] According to the identified abnormal feature type in the blind spot area, the linear feature space data and the nonlinear feature space data are respectively adjusted in weight to generate the blind spot corrected linear feature space data and the blind spot corrected nonlinear feature space data.
[0108] According to the identified abnormal feature type in the blind spot area, the linear feature space data and the nonlinear feature space data are respectively adjusted in weight, which means that the feature dimensions related to the abnormal behavior type in the blind spot area in the two types of feature space data are differentially modified in weight. Specifically, first, based on the identification result of the abnormal feature type in the blind spot area, the feature dimensions related to the abnormal type are determined, such as load fluctuation abnormal feature, equipment start transient current abnormal feature, and load abnormal time series feature; and the feature dimensions are respectively adjusted in weight value in the original feature space.
[0109] For example, the abnormal feature type in the blind spot region is identified as "frequent abnormal device startup" through analysis, and in the linear feature space data after blind spot correction, the weight of the load fluctuation amplitude feature and the abnormal startup frequency feature in the day can be increased; in the nonlinear feature space data after blind spot correction, the weight value of the nonlinear association strength feature between the device startup transient current feature and the load fluctuation feature is correspondingly increased, so as to highlight the coupling relationship between abnormal behavior features.
[0110] For example, the initial weight of the original load fluctuation amplitude feature is 0.2, which can be adjusted to 0.4 after identifying the abnormal feature in the blind spot region, and the initial weight of the device startup frequency feature is increased from 0.15 to 0.3. The above method generates linear feature space data and nonlinear feature space data after blind spot correction. The weight adjustment process improves the effectiveness of abnormal feature identification and eliminates the possible analysis blind spot in the initial fusion analysis process.
[0111] The linear feature space data and nonlinear feature space data after blind spot correction are subjected to secondary fusion processing to output the abnormal behavior identification result of illegal electricity stealing;
[0112] The secondary fusion processing of the linear feature space data and the nonlinear feature space data after blind spot correction means that on the basis of blind spot feature reanalysis and weight adjustment, the fusion analysis process is performed again on the two types of feature space data. The secondary fusion processing can use feature-level fusion or decision-level fusion method, and the feature data space after weight correction enhances the accuracy and stability of user abnormal behavior feature identification.
[0113] Specifically, first, the linear space abnormal features and nonlinear space coupling features after blind spot correction are constructed into a new unified feature space, such as feature-level fusion by weighted summation, weighted average, nonlinear mapping or decision tree method; or decision-level fusion, such as Bayesian fusion, evidence fusion theory method, to calculate the new fusion decision probability and output the final classification and identification result of user abnormal behavior.
[0114] For example, through secondary fusion processing, the new weighted fusion association strength between the user load fluctuation abnormal amplitude and the device running state abnormal feature is calculated, and if the fusion association strength is higher than the predetermined abnormal behavior judgment threshold, it is finally output that the user has obvious load abnormal behavior feature caused by abnormal device startup, which is marked as illegal electricity stealing abnormal behavior.
[0115] The illegal electricity stealing abnormal behavior recognition result generated by the secondary fusion processing can more accurately reflect the user's real electricity consumption behavior abnormal pattern. For example, the user fails to identify abnormal behavior in the initial fusion analysis, and after reanalysis of the blind spot features and adjustment of the weights, the secondary fusion analysis process identifies that the user has continuous abnormal device startup behavior and affects the load characteristics, thereby confirming that the user has illegal electricity stealing behavior.
[0116] The above formulas are dimensionless values calculated, the formulas are obtained by collecting a large amount of data to simulate a formula of the latest real situation, and the preset parameters and threshold values in the formula are set by a person skilled in the art according to the actual situation.
[0117] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another by wired (for example, infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center and the like containing one or more available medium sets. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD) or a semiconductor medium. The semiconductor medium can be a solid state disk.
[0118] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.
[0119] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.
[0120] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. For example, the division of the modules is merely logical function division. There can be another division manner for the actual implementation, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or modules, and can be in electrical, mechanical or other forms.
[0121] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules. They can be located in one place, or can be distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.
[0122] In addition, each functional module in the various embodiments of the present application can be integrated into a processing module, or each module can exist physically independently, or two or more modules can be integrated into one module.
[0123] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that contributes to the technical solutions or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.
[0124] The above description is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
[0125] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.
Claims
1. An anti-theft and violation intelligent identification method based on multi-algorithm fusion, characterized by: The steps include: S1: Obtain the user's real-time electricity consumption monitoring data, and perform feature extraction on the real-time electricity consumption monitoring data to obtain multi-dimensional electricity consumption feature data; S2: Based on the linear mapping algorithm and nonlinear sensitivity algorithm, linear mapping analysis and nonlinear sensitivity evaluation are performed on the multidimensional electricity consumption characteristic data to generate linear feature space and nonlinear feature space data; S3: Perform abnormal pattern recognition processing on the linear feature space data to generate linear space abnormal features; S4: Perform implicit feature correlation identification processing on nonlinear feature space data to generate nonlinear spatial coupling features; S5: Perform fusion correlation analysis on linear spatial anomaly features and nonlinear spatial coupling features to generate fusion feature blind spot area recognition data; S6: Perform blind spot correction analysis based on the fused feature blind spot area recognition data and output the recognition results of illegal electricity theft abnormal behavior.
2. The anti-theft and violation intelligent identification method based on multi-algorithm fusion according to claim 1 is characterized in that: S1, including: Real-time collection of users' power load data and power equipment operating status data; Perform data preprocessing on power load data and power equipment operating status data to remove abnormal and interfering data; The power load data and power equipment operating status data after data preprocessing are subjected to feature extraction processing to extract load fluctuation characteristics, power consumption time series characteristics and power equipment operating characteristics to generate multi-dimensional power consumption feature data.
3. The anti-theft and violation intelligent identification method based on multi-algorithm fusion according to claim 2 is characterized in that: S2, including: Perform linear transformation processing on multi-dimensional electricity consumption feature data based on linear mapping algorithm to generate linear feature space data reflecting the differences in user electricity consumption behavior characteristics; Based on the nonlinear sensitivity algorithm, sensitivity evaluation processing is performed on the multidimensional electricity consumption feature data to determine the nonlinear correlation degree between different electricity consumption features and generate nonlinear feature space data that reflects the implicit dependency relationship of electricity consumption features.
4. The anti-theft and violation intelligent identification method based on multi-algorithm fusion according to claim 3 is characterized in that: S3, including: Perform pattern extraction on linear feature space data to construct an initial abnormal behavior pattern set for identifying user electricity usage behavior; Perform feature matching on the initial abnormal behavior pattern set, and identify feature groups that deviate from normal behavior by comparing the features of each dimension in the linear feature space with predefined standard parameters of electricity consumption behavior; Anomaly clustering is performed based on feature groups that deviate from normal behavior, different types of linear space abnormal behavior features are distinguished, and linear space abnormal feature data is generated.
5. The anti-theft and violation intelligent identification method based on multi-algorithm fusion according to claim 4 is characterized in that: S4, including: The implicit feature correlation evaluation is performed on the nonlinear feature space data, the nonlinear correlation coefficient between the dimensional electricity consumption features is calculated, and the electricity consumption feature dimensions with implicit dependencies are determined; Based on the electricity consumption characteristic dimension with implicit dependency, a nonlinear characteristic dependency network model is established to evaluate the implicit correlation between nodes in the nonlinear characteristic dependency network model. Based on the implicit correlation degree between nodes, feature groups showing stable coupling relationships in nonlinear feature space are identified, and nonlinear spatial coupling feature data are generated.
6. The anti-theft and violation intelligent identification method based on multi-algorithm fusion according to claim 5 is characterized in that: S5, including: Fusing linear space anomaly feature data with nonlinear space coupling feature data to establish a fusion feature data mapping relationship; Based on the mapping relationship of fused feature data, the feature correlation degree is calculated to determine the correlation strength between linear spatial anomaly features and nonlinear spatial coupling features; According to the determined correlation strength, the blind spot area of the fusion feature is identified, the feature range that is not effectively covered in the fusion processing process is determined, and the blind spot area identification data of the fusion feature is generated.
7. The anti-theft and violation intelligent identification method based on multi-algorithm fusion according to claim 6 is characterized in that: S6, including: Based on the fusion feature blind spot area identification data, the blind spot features are re-analyzed to identify the abnormal feature types in the blind spot area; According to the abnormal feature type in the identified blind spot area, the linear feature space data and the nonlinear feature space data are weighted respectively to generate the linear feature space data and the nonlinear feature space data after the blind spot correction; The linear feature space data after blind spot correction and the nonlinear feature space data are subjected to secondary fusion processing to output the abnormal behavior recognition results of illegal electricity theft.